．單位首頁
．系所簡介
．系主任
．強力師資
．研究所
．課程概況
．各類學程
．期刊論文
．招生資訊
．演講專區
．榮譽榜
．系學會
．系友天地
．就業訊息
97年2月22日(五) 2:00 ~3:00 P.M.
演講者姓名： 王清雲 教授
演講者服務單位： Division of Public Health Sciences, Fred Hutchinson Cancer Research Center
Instrumental Variable Estimation in Logistic Regression With Covariate Measurement Error When Repeated Surrogates Are Not Available
Abstract
We consider logistic regression when covariates are measured with errors. When a gold standard measurement for an exposure variable is not available for any of the study subjects, a common approach is to use repeated surrogate variables. However, in some real examples, there is only one surrogate variable available for each subject. To deal with an identifiability issue, a general approach is to search for a variable that is relevant to the unobserved covariates. The variable, that can provide useful information about the unobserved covariate, is often considered as an instrumental variable. In this article, we investigate estimation for the logistic regression when there is a single unbiased surrogate and a single instrumental variable. In addition to the estimation of the regression parameters, an important issue is to estimate the standard error of the measurement error. The result can be applied to studies when repeated surrogates for an unobserved covariate are not available. Results from a simulation study are presented to show the finite sample performance of the method.
Keywords: conditional score, corrected score, error-in-variable, functional methods.
